Experimenting with PlanetScale. Coming from Neon. I knew it was going to be fast, even at p99, and it is.
The interesting thing is that they don't separate compute from storage as the other innovations in the postgres world are doing. In return it's reliable, fast, and hands-off. That's what I want in an OLTP database, roughly in order. Reliable, fast, hands-off.
Neon, Aurora, etc make different tradeoffs. With those you can scale to zero, scale storage but not compute, and scale compute to meet demand instead of over-provisioning. I have other projects for which that's perfect, but for a real user-facing webapp I want reliable, fast, and hands-off.
Worth it even if it costs more. (Probably won't though.)
Thanks to @ayrton for setting me up with early access to the $5 kick-the-tires plan.
5
98
10,826
The uptime of major LLM providers is lower than would be acceptable for any other service, but still impressively high given the complexities of batching, scheduling, etc.
↖️ Anthropic
↗️ OpenAI
↙️ Google
↘️ Xai
136
It appears that Americans do not care what celebrities think about politics. In 2028 we'll see many fewer celebrity endorsements. Candidates will de-emphasize them too.
1
113
Just gave an invited talk about visx.io at the @WolframResearch Technology Conference. Thank you for the great response!
2
1
176
This is good. You can raise your score a little with SAT prep but you can't fake a 790 (the average math score for incoming students). This makes it an objective measure which can highlight people whose backgrounds are different from the usual applicants. mitadmissions.org/blogs/entr…
Thanks to the @ai_guild for the 4th #datalift event. Good overview from @borisschauerte on data science at @BMWGroup.





